Deep Learning and Neural Networks

As the video you watched in the previous section (What is Aritificial Intelligence?) makes clear, artificial intelligence is distinguished from other computational methods by its use of “deep learning.” Unlike “hard-coded” algorithms that instruct a computer to carry out a specified series commands or allow it to choose an action based on information previously supplied to it (such as actual chess matches), machine learning algorithms use statistical inference to enable a computer to predict outcomes and revise those predictions in response to feedback. Over time, the computer “learns” from repeated feedback to make better and better predictions.

We need to keep those quotation marks around the word learns because this mode of learning is different from human learning. Much of the terminology associated with artificial intelligence suggests an analogy with human mental processing, and that’s probably fine as long as we keep in mind that it’s only an analogy. But as we saw from the video in the previous section, the history of artificial intelligence in computing is bound up with efforts to emulate or compete with human intelligence, and one consequence of this effort seems to be a tendency, on the part of some researchers and corporations, to deliberately obscure the difference between them. We’ll return to this tendency in the section on AI and ethics.

The second video in William J.B. Mattingly’s series explains a core tool of artificial intelligence named for the human anatomical feature it’s designed to emulate: the neural network. Again, keep in mind that the structure and behavior of the neural networks used in computing is not the same as the structure and behavior of the actual networked neurons in the human brain.